| name | ormus-illustrator |
| description | Generate AI illustrations and videos. Triggers when user wants images, videos, or animations created. Supports image-to-video and text-to-video generation. |
| triggers | ["illustrat","generate image","create image","make an image","draw","visual","picture of","artwork","graphic","render","midjourney","pixel art","animation","sprite","animate","video","generate video","image to video","text to video","animate image"] |
| version | 4.1.0 |
Ormus Illustrator
Multi-provider AI illustration engine with an LLM-first philosophy.
"LLMs are the chill, reliable grown-up in the room."
The Core Philosophy
Image generators wake up each morning with a new personality. Even the best ones struggle with consistent personal style. Midjourney has taste, but also a temper.
The solution: Use Claude as a thinking partner before touching any image tool.
| Role | Responsibility |
|---|
| Claude | Structure, consistency, prompt refinement |
| Image Tool | Aesthetics, visual execution |
| You | Taste, curation, direction |
This creates stability in an inherently chaotic process.
Phase 0: Think Before You Generate
Before jumping into any provider, have a conversation with Claude about:
1. Context Building
Ask yourself (or let Claude help you articulate):
- What am I creating? (illustration, icon, character, scene?)
- Who is it for? (children, developers, myself?)
- Tone — playful, serious, ethereal, scientific?
- Colors — specific palette or mood-driven?
- References — what existing work feels "right"?
2. Personal Design Scaffold (Optional but Powerful)
Write down how you think when designing:
- What do you look at first?
- How do you choose a direction?
- What do you avoid?
- How do you explore ideas?
Once Claude knows your logic, prompts feel surprisingly aligned.
3. Example Conversation Opener
I'm creating illustrations for [project].
Target audience: [who]
Feeling I want: [adjectives]
Colors I'm drawn to: [colors]
Things I want to avoid: [anti-patterns]
References that feel right: [links or descriptions]
Help me think through this before we generate anything.
The Positive Loop
This is where the magic happens:
┌─────────────────────────────────────────────────────┐
│ │
│ Claude → Prompt → Image Tool │
│ ↑ │ │
│ │ ↓ │
│ Tell Claude ← Mood Board ← Output │
│ "we evolved" │
│ │
└─────────────────────────────────────────────────────┘
- Chat with Claude about what you're making
- Share references from your mood board
- Let Claude build prompts that match your logic
- Run them in the appropriate provider
- Take good results → add to mood board
- Tell Claude "Look, we just evolved the style!"
After a few rounds, your style becomes surprisingly stable.
Mood Board Workflow
Your mood board is your visual anchor.
Collect things you love:
- Colors and gradients
- Textures and patterns
- Photography/illustration styles
- Small visual cues that feel "right"
Keep it clean. A cluttered board = unclear direction.
Storage suggestion: ~/projects/[project]/mood-board/
Evolve it. As you generate good outputs, add them back. The board grows with your style.
When This Skill Activates
- User says "illustrate", "draw", "create an image", "generate a picture"
- User requests visual content for any project
- User needs graphics, artwork, or rendered images
- User mentions "midjourney" for artistic/hand-drawn styles
- User wants "pixel art", "animation", "sprite", or "animate"
Available Providers
| Provider | Best For | Personality | Speed | Cost |
|---|
| Gemini | Botanical, scientific | Precise, literal | Fast | Free (quota) |
| Replicate | Icons, vectors, video | Versatile, predictable | Medium | ~$0.003-0.07 |
| Midjourney | Artistic, hand-drawn, childlike | Has taste, has temper | Slow | Subscription |
| PixelLab | Pixel art, game animations | Consistent, game-focused | Fast | Credits-based |
Provider 1: Gemini (Default)
cd ~/.claude/plugins/ormus-illustrator && python3 generate.py "PROMPT" --style STYLE --name FILENAME
Style presets: botanical, scientific, alchemical, minimal, none
Auth: ~/google-cloud-sdk/bin/gcloud auth application-default login
Provider 2: Replicate (Images & Video)
cd ~/.claude/plugins/ormus-illustrator/providers && python3 replicate.py "PROMPT" --model MODEL --name FILENAME
Image Models
| Model | Best For |
|---|
flux-schnell | Fast drafts (~$0.003) |
flux-pro | High quality |
recraft-v3 | Icons, text, logos |
recraft-svg | Clean SVG vectors |
ideogram-v2 | Perfect text in images |
List image models: python3 replicate.py --list-models
Video Models (NEW)
| Model | Type | Best For |
|---|
svd | Image→Video | Animate a still image (~$0.07) |
animate-diff | Image→Video | Motion with prompt control |
i2vgen-xl | Image→Video | High quality animation |
minimax | Text→Video | Generate video from description |
hunyuan | Text→Video | High quality text-to-video |
ltx-video | Text→Video | Fast text-to-video |
liveportrait | Animation | Animate faces/talking heads |
List video models: python3 replicate.py --list-video-models
Video Generation Examples
Image-to-Video (animate a still image):
python3 replicate.py --video --model svd --image my-character.png --name character-animated
Text-to-Video (generate from description):
python3 replicate.py "a robot walking through a forest" --video --model minimax --name robot-walk
Animate with motion prompt:
python3 replicate.py "gentle breathing motion" --video --model animate-diff --image portrait.png --name breathing
Provider 3: Midjourney (Discord Automation)
Best for: childlike, hand-drawn, artistic, crayon-style illustrations.
cd ~/.claude/plugins/ormus-illustrator/providers && python3 midjourney.py "PROMPT" --name FILENAME
Single image:
python3 midjourney.py "childlike crayon drawing of a happy sun, wobbly lines, bright colors --v 6.1 --s 200" --name happy-sun
Batch generation (JSON file):
python3 midjourney.py --batch prompts.json --output ~/project/assets/
Batch JSON format:
[
{"name": "sun", "prompt": "childlike crayon sun --v 6.1"},
{"name": "heart", "prompt": "hand-drawn heart, marker style --v 6.1"}
]
Midjourney parameters:
--v 6.1 - Latest version (default)--niji 6 - Anime style--style raw - Less stylization--s 0-1000 - Stylization amount (lower = more literal)--ar 1:1 - Aspect ratio
First-time setup:
- Run the script - browser window opens
- Log into Discord manually
- Session is saved for future use
Quick Decision Guide
Images
| Want | Use |
|---|
| Fast draft | Replicate (flux-schnell) |
| SVG/vectors | Replicate (recraft-svg) |
| Text in image | Replicate (ideogram-v2) |
| Childlike/hand-drawn | Midjourney |
| Botanical/scientific | Gemini |
| Artistic/ethereal | Midjourney |
Video & Animation
| Want | Use |
|---|
| Animate still image | Replicate (svd, animate-diff) |
| Text to video | Replicate (minimax, ltx-video) |
| Talking head / face | Replicate (liveportrait) |
| Pixel art animations | PixelLab |
| Walk cycles/sprites | PixelLab |
| Retro game art | PixelLab |
Output Locations
- Default:
~/.claude/plugins/ormus-illustrator/output/
- Per-project: Use
--output "PATH" flag
Troubleshooting
| Error | Provider | Solution |
|---|
| 401 Unauthorized | Gemini | gcloud auth application-default login |
| 429 Rate Limited | Any | Wait 1-2 minutes |
| Discord login needed | Midjourney | Log in manually in browser window |
| Browser not found | Midjourney | playwright install chromium |
Example: La Gran Obra Childlike Style
For hand-drawn, crayon-like, blob-shape illustrations:
python3 midjourney.py "childlike crayon drawing, wobbly organic blob shape, bright yellow color, hand-drawn imperfect lines, no background, simple joyful character, marker texture --v 6.1 --s 150 --ar 1:1" --name yellow-blob --output ~/projects/01-ACTIVE/la-gran-obra/public/assets/shapes/
Style keywords for childlike aesthetic:
- "childlike crayon drawing"
- "hand-drawn imperfect lines"
- "wobbly organic shapes"
- "marker texture"
- "finger-paint style"
- "simple joyful"
- "no background" or "transparent background"
Provider 4: PixelLab (Pixel Art & Animations)
Best for: pixel art animations, sprite sheets, walk cycles, idle animations, retro-style game art.
cd ~/.claude/plugins/ormus-illustrator/providers && python3 pixellab.py "ACTION" --image REFERENCE.png --name OUTPUT
Generate animation from reference:
python3 pixellab.py "walking" --image turpi.png --name turpi-walk --size 64 --frames 4
Generate idle animation:
python3 pixellab.py "breathing idle" --image turpi.png --name turpi-idle --frames 8
Generate waving animation:
python3 pixellab.py "waving hello" --image turpi.png --name turpi-wave --size 32
Parameters:
| Flag | Description | Default |
|---|
--image/-i | Reference image (required) | - |
--name/-n | Output name prefix | animation |
--size/-s | Output size (32, 64, 128) | 64 |
--direction/-d | Character direction (north, south, east, west, etc.) | south |
--frames/-f | Number of frames (2-20) | 4 |
--seed | Random seed for reproducibility | - |
--keep-bg | Keep background (don't remove) | - |
--output/-o | Output directory | default |
Estimate skeleton from image:
python3 pixellab.py --skeleton turpi.png
List animation templates:
python3 pixellab.py --list-templates
Available templates: backflip, breathing-idle, cross-punch, die-and-revive, falling, hook-punch, idle-to-crouch, idle, jump-kick, jump, run, slash, slide, smash-with-sword, spin-kick, walk, wave, wings-flap
API Key: Stored in script or set PIXELLAB_API_KEY environment variable.
MCP Integration: PixelLab MCP server available at https://api.pixellab.ai/mcp (configured in project)
Gentle Iteration (No Need to Grind)
Early results might feel rough — totally normal.
But as the loop continues:
- Your prompts become sharper
- The tool begins to "understand" your vibe
- Your mood board gains personality
- A unique style emerges
You'll notice something special:
- Claude handles structure
- The image tool handles aesthetics
- You handle taste
The Anti-Patterns
| Don't | Why |
|---|
| Jump straight to generation | You lose the thinking phase |
| Overload mood boards | Cluttered board = unclear direction |
| Expect instant perfection | Style emerges through iteration |
| Ignore good "accidents" | Sometimes the tool surprises you — save those |
| Fight the tool's personality | Work with its strengths, not against them |
Final Wisdom
This workflow is not about being technical. It's about:
- Reducing guesswork in an inherently chaotic process
- Building a stable creative backbone with Claude as the consistent partner
- Letting AI understand your taste through conversation
- Growing your style slowly, naturally, one prompt at a time
It's simple, really. Just a conversation between you and your tools.
No pressure. No heavy theory. Just a path that helps your visual voice grow.
v4.0.0 | Philosophy integrated from LLM-first image generation workflow
